Understanding how students reason about code is essential for providing tailored scaffolding in computer science (CS) education. Prior work has used think-aloud protocols with the Structure of the Observed Learning Outcomes (SOLO) taxonomy to examine students' code comprehension and programming levels. However, analyzing such data is labor-intensive and requires expert judgment. Recent advances in large language models (LLMs) offer a promising avenue for scaling this analysis, though their reliability for fine-grained coding remains uncertain. To address this gap, our study investigates the extent to which GPT-5 and 4o can classify SOLO levels and identify code-comprehension strategies from think-aloud transcripts of 27 high-school students working on block-based and text-based tasks. Results show modest alignment with human ratings for SOLO, with one-shot prompting improving agreement over zero-shot, though distinctions between adjacent lower levels (e.g., Prestructural 1 vs. 2) remained difficult. Strategy detection demonstrated stronger performance, achieving accuracies of 75–77% (block) and 62–67% (text), particularly for surface-visible strategies such as 'walkthroughs', 'control-structure identification', and 'pattern recognition', but weaker for less frequent, abstract, meta-cognitive strategies such as 'strategizing' (planning an approach) or 'thoroughness' (systematically checking work). These findings highlight both the potential and the limitations of using GPT-5 and 4o to analyze think-aloud data. While this work represents an initial step, with plans to examine more models, our preliminary results indicate that a human-in-the-loop approach is essential to ensure reliability and interpretive depth. Future work will extend this evaluation to other LLMs to better understand their role in supporting instructional decision-making.

Understanding how students comprehend code is essential for designing effective instructional support in computer science (CS). While prior studies have often relied on written responses, few have examined students' reasoning processes through think-aloud data. In this study, we analyzed the verbal reasoning of 27 high school students as they completed block-based and text-based code comprehension tasks targeting loops and conditional statements. Using an adapted SOLO taxonomy framework, we found that most students were classified at lower levels, with performance declining as they transitioned from block-based to text-based code. Students' strategy use, informed by prior work on code comprehension, showed that walkthroughs and identifying program structures were the most common approaches. Text-based tasks more often led students to use pattern-recognition strategies, such as interpreting operators or identifying numerical patterns, whereas block-based tasks occasionally prompted them to articulate broader problem-solving approaches. Overall, these findings demonstrate the value of applying the SOLO taxonomy to evaluate students' programming levels and highlight how programming modality impacts both the depth of understanding and the strategies students employ during code comprehension.

Question-asking is a crucial learning and teaching approach. It reveals different levels of students' understanding, application, and potential misconceptions. Previous studies have categorized question types into higher and lower orders, finding positive and significant associations between higher-order questions and students' critical thinking ability and their learning outcomes in different learning contexts. However, the diversity of higher-order questions, especially in collaborative learning environments. has left open the question of how they may be different from other types of dialogue that emerge from students' conversations, To address these questions, our study utilized natural language processing techniques to build a model and investigate the characteristics of students' higher-order questions. We interpreted these questions using Bloom's taxonomy, and our results reveal three types of higher-order questions during collaborative problem-solving. Students often use "Why", "How" and "What If' questions to I) understand the reason and thought process behind their partners' actions: 2) explore and analyze the project by pinpointing the problem: and 3) propose and evaluate ideas or alternative solutions. In addition. we found dialogue labeled 'Social'. 'Question - other', 'Directed at Agent', and 'Confusion/Help Seeking' shows similar underlying patterns to higher-order questions, Our findings provide insight into the different scenarios driving students' higher-order questions and inform the design of adaptive systems to deliver personalized feedback based on students' questions.

With the increasing prevalence of large language models (LLMs) such as ChatGPT, there is a growing need to integrate natural language processing (NLP) into K-12 education to better prepare young learners for the future AI landscape. NLP, a sub-field of AI that serves as the foundation of LLMs and many advanced AI applications, holds the potential to enrich learning in core subjects in K-12 classrooms. In this experience report, we present our efforts to integrate NLP into science classrooms with 98 middle school students across two US states, aiming to increase students' experience and engagement with NLP models through textual data analyses and visualizations. We designed learning activities, developed an NLP-based interactive visualization platform, and facilitated classroom learning in close collaboration with middle school science teachers. This experience report aims to contribute to the growing body of work on integrating NLP into K-12 education by providing insights and practical guidelines for practitioners, researchers, and curriculum designers.

Intelligent virtual agents have tremendous potential for facilitating collaborative learning by modeling and reinforcing desirable collaborative practices. Despite recent work in this area, the extent to which intelligent virtual agents can facilitate improvements in the collaborative behavior of children is largely unknown. This study employed a wizard-of-oz study design and investigated elementary children's collaborative behavior after interacting with virtual agents. These agents model exploratory talk for upper elementary school dyads, such as asking higher-order questions and listening to their partners. The findings uncover associations between elementary learner dyads' positive changes in collaboration after agent interventions, the dyads' affective reactions to interventions, and their attentiveness to the agents. Our results also reveal associations between positive changes in collaboration and the timing of interventions: for example, earlier interventions had a higher occurrence of positive changes, and positive changes in collaboration typically happened within five seconds of interventions. The results suggest ways in which intelligent virtual agents may be used to promote effective collaborative learning practices for children.